Executive Communication and Decisions

A fictional healthcare-software company’s CTO brought forty accurate slides about a claims-processing engine to the executive committee. They covered architecture, vendors, migration phases, staffing, and risks, but did not connect those facts to a recommendation and an exact request. The committee asked her to return with a decision it could assess. Two weeks later, a security incident exposed a second communication problem: the committee learned of the company’s required notification before receiving an internal update from the CTO, who had been waiting for a complete explanation. The notification itself was not evidence that replacement would have prevented the incident. This article examines decision documents, proportionate review, oversight reporting, and early communication under uncertainty. The examples are invented; the guidance comes from public appraisal standards and published company practices.

What a decision needs: six things

For an investment request, a useful organizing structure is the problem, alternatives, economics, risks and delivery arrangements, recommendation, and exact authorization sought. These are the author’s prompts, not a complete specification for every executive document. The supporting evidence must fit the decision and its stage. The UK Treasury’s five case model provides a related way to examine a proposal through strategic, economic, commercial, financial, and management perspectives. The following mapping keeps their distinct responsibilities visible.

Decision document	Five Case Model perspective	What to establish	Gap in the fictional deck
Problem and baseline	Strategic	Objectives, case for change, and expected business as usual	Architecture described, but objectives and future baseline not connected
Alternatives and value	Economic	Feasible options, costs, benefits, uncertainty, and reasons for preference	Vendor information without a comparable options appraisal
Affordability	Financial	Funding, expenditure and receipts over time, and budget constraints	A quote without a complete funding schedule
Sourcing, risk, and implementation	Commercial and management	Supplier capability, procurement and contract terms, risk allocation, staffing, change, benefits, and evaluation plans	Plans and risk labels present, but assumptions and commitments untested
Recommendation	Judgment across the cases	Preferred option and remaining conditions	Absent
Request	Authorization for the present stage	Decision, scope, limit, authority, and next review	Absent

Business as usual (BAU) describes the expected future if the proposed intervention is not implemented. It includes continuing operation, maintenance, obligations, costs, benefits, and risks. It is not necessarily doing nothing. A do-minimum option just meets the proposal’s objectives without extra benefits. Keeping these separate helps reviewers compare a proposed improvement with a realistic baseline and a modest intervention. The five cases also develop together, with detail proportionate to the stage, cost, complexity, and risk. Early exploration does not require completed procurement, but it does need enough commercial, funding, and delivery information to judge that exploration.

Generate alternatives before choosing a favorite. Assess mandatory conditions with evidence and distinguish pass, fail, and unverified. Among eligible options, compare value and trade-offs; an option can satisfy the conditions and still be less attractive than another. Under Green Book guidance, BAU remains in the shortlist as a benchmark even when it does not meet the objectives or critical success factors. That comparison does not authorize selecting a noncompliant baseline. The article on technology strategy develops the argument from goals to choices. The article on budgets and total cost explains comparable costs, including how to use evidence when correcting optimism bias. The public model concerns social value as well as public funding; a company borrowing its structure must state whose outcomes and costs its analysis includes.

The economics an executive can act on

A benefit-cost ratio, total cost, or payback period describes part of an appraisal. None settles the choice alone. Green Book appraisal combines value measures with unmonetized effects, uncertainty, distributional impacts, and financial requirements. Sensitivity analysis varies specified assumptions while keeping the comparison and other inputs explicit. For a cash model, vary implementation cost, net savings, timing, or discount assumptions, then examine plausible joint scenarios. This is related to sensitivity analysis in decision scoring, where weights, scoring anchors, and measured inputs affect rankings. The two applications use different models. Switching values identify boundaries within a specified comparison; a boundary for three-year payback is different from one for total net value or preference between options.

For a simple illustration, use arbitrary currency units and five end-of-year cash flows relative to a common BAU, with both options meeting required outcomes. Option A costs 6 now and saves a net 2 each year. Its undiscounted cumulative cash flow reaches zero at year three. At an illustrative discount rate of 10%, its five-year NPV is −6 + Σ(2 / 1.1^t) for t = 1 to 5, or about 1.582. Option B costs 1 now and saves a net 0.8 each year, giving NPV about 2.033. A meets the three-year payback test but B has greater net cash value under these assumptions. This is a corporate cash illustration, not the Green Book’s social discount rate or a complete social appraisal. Show the underlying schedule and important noncash outcomes before recommending either option.

A concise request with evidence behind it

Amazon’s 2017 shareholder letter describes its use of six-page narratives, silent reading, and repeated revision with colleagues. It explains that developing a strong memo may take a week or more, including time away from the draft, and that writing is a team activity. That elapsed time is not a fixed requirement for one senior person’s labor. Narrative can expose missing connections between claims and evidence, but it can also conceal weak reasoning. A slide deck can carry a coherent argument. Review the reasoning rather than treating either format as proof of quality.

For this proposal, the CTO could begin with a concise request: the decision needed now, its scope and funding limit, and the conditions for proceeding. Supporting material would then explain the problem and BAU, eligible alternatives, comparative value and affordability, sourcing, delivery risks, and review arrangements. A short summary helps a busy reader find the request; tables, diagrams, calculations, and a longer narrative can provide the evidence. Their length should follow the questions the approver must resolve. A statement of why an alternative was less attractive is more informative than calling every rejected option a failure.

GitLab’s guidance encourages authors to explain why a change is being made and preserve that reasoning for future readers. For the CTO, that meant making the preferred option’s assumptions and trade-offs inspectable, so the committee could challenge them or ask for specific missing evidence.

Match review effort to consequences and reversibility

The two-way door metaphor draws attention to how a decision could be reversed at acceptable cost and within a useful time. GitLab and Amazon use reversibility to support proportionate review and faster iteration. It is one consideration alongside potential harm, uncertainty, obligations, and authority. Reverting a configuration does not recover information already disclosed or money already paid. A budget allowance also does not confer data-access, procurement, or production-release authority.

Decision	Reversal questions	Evidence and review	Authority in this fictional company
Compare two engines in an isolated 90-day test	Can processing stop and test copies be removed without production effects or a purchase commitment?	Approved test data, isolation, spending cap, success and stop criteria, exit terms	Committee authorizes the cross-functional experiment; data and security approvals remain separate
Change release cadence	Can the schedule change back, and what effects of intervening releases would remain?	Service obligations, testing capacity, change risk, and recovery arrangements	Service and release owners within their written delegations
Sign a five-year replacement contract	What termination rights, costs, dependencies, and data-exit arrangements apply?	Full proposal appropriate to the commitment; commercial and funding evidence	Authorized procurement and funding route, supported by the sponsor within that role’s mandate
Retire the current engine	Which records, capabilities, and staff arrangements can be restored, and by when?	Validated replacement, retention and access requirements, continuity and recovery evidence	Named retirement authority with required business and control approvals

The table describes this company’s assignment, not a hierarchy inferred from the size of the document. Its committee had authority to fund the cross-functional test; the CTO did not have sole authority to commit all affected teams or approve their data use. A different company might validly delegate a similarly bounded experiment. GovS 002 calls for defined decision makers, consultation, and authorization criteria at gates. The article on stakeholders and sponsors explains how to establish those mandates. Assess whether the proposed rollback has been demonstrated: missing evidence means reversibility is unverified, not that the action is necessarily irreversible.

Reporting: a flow of information timed to decisions

Oversight reports have several purposes: service performance, controls, emerging risks, investment progress, and continuing business justification. The UK digital leadership framework calls for performance-based insights and recommendations to the board. An investment report should revisit the expected benefits and rationale, while an operational report may need to show persistent service or control conditions even when no new decision is requested. A report can be useful without asking to change the budget.

A compact item can state its current position against a baseline or limit, what changed, what remains uncertain, and the action or awareness needed. Retain serious unresolved conditions even when they have not changed. Give future decisions enough review time for their evidence, dependencies, and urgency; two meetings’ notice is one possible planning choice, not a universal rule. Material events need escalation between meetings. The benefits realization entry distinguishes forecasts, observed changes, and demonstrated benefits. Where a metric and a frontline account differ, check their populations, periods, definitions, and coverage before concluding that either is wrong.

Bad news: early, with uncertainty made explicit

Waiting for a complete root-cause explanation had delayed useful internal awareness. GitLab’s underperformance guidance illustrates early escalation to a manager in a people-management context; it does not define incident-notification duties. For a production incident, agree severity thresholds, response roles, internal escalation routes, and applicable customer or regulatory notification responsibilities in advance. Executive awareness, incident coordination, and formal external notification are related activities with different audiences and owners. An update to the CEO must not become a prerequisite that delays authorized response or required notification.

A concise initial update can state what is confirmed, what is suspected or unknown about the impact, who is coordinating the response and what is underway, and what help is needed. Include the observation time and next update time. A suspected breach should not be announced as an established fact merely to sound decisive. Escalate using the agreed severity and obligation criteria rather than waiting for certainty; an urgent call may precede a written note. For a likely delivery miss, explain the affected commitment, forecast range, reasons, recovery options, and requested decision. The objective is timely, usable information, not ensuring that one executive always hears first.

Write the decision down

A decision record preserves the authorization, reasoning, evidence, rejected alternatives, conditions, and review triggers. It reduces avoidable reargument but does not guarantee agreement or execution. GitLab’s decision guidance directs people seeking a change to the DRI who can make it. Follow the currently authorized change route, which may include emergency suspension and escalation rather than repeating the original meeting. Record the superseding decision and reason. Changing a decision about future work cannot undo all consequences of work already performed.

The second meeting

The revised proposal requested committee authorization for an isolated comparison of two engines over at most 90 days, with an illustrative spending cap of 0.2 million in the company’s currency and named staff commitments. No production claim would be paid or changed by the trial. The data owner and security lead had to approve the test dataset and environment before access began; procurement had to confirm exit terms and the absence of an automatic purchase commitment. The budget allocation alone satisfied none of those conditions. The summary identified those prerequisites and linked the supporting evidence.

The trial would test whether either candidate met the agreed claim-calculation rules, representative volume and exception cases, audit-trace requirements, and export and restoration tests. Business owners supplied reference results and coverage, including high-risk cases. An unexplained critical calculation difference, unauthorized data access, production effect, or projected cap breach would stop the affected test and trigger investigation. The evaluation owner would report failures, remaining gaps, and limits of extrapolating from one claims line. The committee could extend a properly authorized investigation, change scope, or stop; reaching day 90 would not automatically approve replacement.

The options comparison retained continued operation and maintenance as BAU and a do-minimum remediation option. Mandatory failures and unverified conditions were separated from preferences among eligible options. Historical incident costs explained the problem and informed future forecasts, but already incurred costs were excluded from incremental future appraisal. The analysts compared expected future losses under maintained BAU, remediation, and replacement, including residual and migration risks; they did not assume replacement would prevent every incident. TCO, funding schedules, value measures, and sensitivity scenarios accompanied the recommendation. The pilot request separately explained which uncertain claims the test could resolve and why that learning was worth its cost.

The committee authorized the experiment subject to its prerequisites, spending cap, stop rules, and a documented evidence review. Its approval did not authorize production rollout or a five-year purchase. The record named the evaluation and resource owners and the separate control approvals. For incidents, the company adopted an illustrative one-hour internal notification target for its high-severity category, with urgent escalation sooner when needed and update intervals agreed by the response lead. Formal notifications followed their own applicable requirements and authorized owners. Scheduled board reports retained current exposure as well as changes and flagged forthcoming decisions according to the review time they needed.

The revised package kept useful architecture diagrams and calculations as supporting material. It separated the experiment from the later procurement and transition decisions, while explaining how evidence would connect them. The improvement was a clear, authorized request and an explicit response to uncertainty. Document length and meeting speed were not evidence that the chosen engine would succeed.

Suppose the fictional CTO subsequently proposes AI-assisted claims triage. The request must distinguish funding an evaluation from authorizing its use in claims decisions. Start with the current triage workflow as BAU, including staff review, exceptions, cost, and errors, and compare improving that workflow with adding AI assistance. The proposal should name the permitted task and data scope, evaluation owner, business decision owner, and separate data, security, and release approvals. Representative cases and known difficult exceptions need reference judgments supplied by authorized domain reviewers. Set acceptable error and escalation criteria before testing, and report where evidence is missing or results differ between relevant claim groups. Approval for a test does not authorize automated claim determinations or wider data access.

The AI investment case also needs the costs of human review, integration, model use, monitoring, and fallback operation alongside any expected time saving. Show sensitivity to workload, review effort, error rates, and changing model behavior rather than treating a demonstration as a dependable saving. A risk update should explain the affected task, what is known and uncertain, the owner of the response, and the exact decision needed: continue a bounded evaluation, change scope, suspend use, or seek approval for a defined rollout. Operating readiness includes a tested route to staff review or the existing triage process when the AI service is unavailable or its output is unsuitable. A material model or workflow change reopens the relevant evaluation and approvals. This makes AI uncertainty part of the same authorized decision process used for the engine investment.

What a reviewer should ask

A document for executives, or a leader’s reporting practice, can be reviewed with a short list, each item tracing to something above.

• Does the opening identify the decision needed now, its scope, limit, authority, and unresolved conditions?
• Is future BAU explicit and distinct from do-minimum intervention? Are historical costs distinguished from future incremental costs?
• Are mandatory conditions tested separately from preferences, with reasons for keeping, deferring, or rejecting options?
• Do the economics show comparable outcomes, costs, funding, and uncertainty? Does each switching value name its comparison and assumptions?
• Do risks have owners, response and acceptance arrangements, and escalation triggers?
• Have reversal time, cost, evidence, and lasting effects been assessed alongside impact and actual approval rights?
• Does reporting retain the current baseline, forecast, unresolved exposure, and relevant decisions, with event-based escalation?
• Do incident updates distinguish confirmed facts, uncertainty, response ownership, and next updates without delaying other required communications?
• Are authorization and reasoning recorded, with current change routes and conditions for stopping, revisiting, or superseding the decision?

The CTO’s first deck contained useful facts but left the decision unformed. Her delayed incident update left the committee without information it needed while the situation developed. Clear requests, proportionate evidence, valid authority, and timely updates made those gaps addressable. Each communication still needed scrutiny against what the organization could know and do at that stage.

References

• HM Treasury, The Green Book: Central Government Guidance on Appraisal, 2026 edition: chapters 3–6 and 9, Tables 2–3, and Box 1.
• UK Government Functional Standard GovS 002: Project delivery, version 2.1, 1 September 2025: principle 3 and the gate and decision-point clauses.
• Amazon, 2017 Letter to Shareholders, section on six-page narratives.
• Amazon, Leadership Principles.
• GitLab Handbook, Values (decision-making and transparency sections).
• GitLab Handbook, Underperformance.
• UK Government Digital and Data Profession Capability Framework, Chief digital and information officer, published 29 August 2025.